graph-engineering

A guide for designing and reviewing durable workflows made from Codex project graphs. These graphs describe how agents, supervisors, reviewers, approval steps, and evidence should work together.

In plain words
What is it for?
Use it to create, audit, or evolve governed agent workflows, including routing work to sub-agents, adding review loops, and defining human approval points.
Why use it?
It helps decide when a repeatable workflow needs explicit controls and how much oversight is appropriate for its risk.

Skill for Claude CodeCodex

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add skills/luxiaolei/graph-engineering/graph-engineering
Any agent
npx skills add luxiaolei/graph-engineering --skill graph-engineering
Clone the repo
git clone --depth 1 https://github.com/luxiaolei/graph-engineering

Made for: Claude Code, Codex.

Per session 75 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,881 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.00075 $0.01881
Opus 5 $0.00037 $0.00941
Sonnet 5 $0.00015 $0.00376
Haiku 4.5 $0.00007 $0.00188

Measured 2d ago against content hash a9942747dc66, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

graph-engineering scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.

The scan reads SKILL.md. This mod also ships 3 executable files (scripts/generate_managed_supervisor_prompt.py, scripts/integrate_supervisor_guidance.py, scripts/validate_project_graph.py), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

skills/graph-engineering/SKILL.md · 154 lines

How it starts

The opening of the file, as written. The whole thing — 154 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Graph Engineering

Design the smallest project-local Graph contract that matches the workflow's actual risk. Do not build a graph runtime and do not execute the target business workflow.

Select the mode and risk tier

  • Create establishes a new durable or repeatable workflow.
  • Audit checks an existing .codex/graphs contract without editing it.
  • Evolve changes a Skill, Agent, or Graph from measured execution evidence.

Derive Graph necessity from the action facts before asking any question. A user-authorized, one-time, bounded, reversible local task with actual_authority_effect=none is NO_GRAPH: proceed under repository rules and do not ask to create or evolve a Graph. Durable, repeatable, or authority-changing work may justify a Graph. When a Graph is justified, choose the lowest sufficient tier:

Tier Use when Extra control plane
workflow-lite/v1 or no profile reversible local workflow; low blast radius base Graph only
reviewed-batch/v1 material code or governed-artifact batch one immutable-batch review, repair limit 0 or 1, one terminal record
authority-critical/v1 data admission, sealed evidence/OOS consumption, broker or trading authority, irreversible publication, secrets/permissions, or comparable blast radius closed admission pack, exact bounded review, serial integration, terminal seal

governed-implementation/v1 is validation-only legacy compatibility; never use it in a new proposal or template. Do not select an authority-critical profile merely because a task is complex, long-running, or important.

Read discovery before create/evolve, the contract before writing Graph files, Codex mapping before defining delegated roles, governance for evolve work or external side effects, and multi-repo guidance when more than one repository is in scope.

Read the full file on GitHub · 154 lines

Files

What ships with it

29 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen · 154 lines · 75 tokens per session scan A a9942747dc66

Subscribe to this mod's changes

graph-engineering is a skill published in the GitHub repository luxiaolei/graph-engineering (2 stars, last pushed 1mo ago), licensed Apache-2.0. It adds 75 tokens to every session and 1,881 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.

Related

Other skills, from other repositories

graph-engineering-workflow

Plan and verify bounded agent graphs for complex coding work. Use when a coding agent needs to coordinate independent implementation, research, verification, auditing, isolated writers, merge ownership, or targeted repair loops. Prefer a single loop for small or genuinely sequential tasks.

Thanarak-q/graph-engineering-workflow · 56 tokens

adversarial-review

Use on a document that matters, that has been revised many times, that its author believes is complete — isolated attackers try to kill its conclusions before reality does.

InjayTseng/graph-engineering-on-research · 37 tokens

consultant-roundtable

Use for a decision with no right answer in public data — pricing, timing, build vs buy. Two-round Delphi - isolated consultants take positions, see the anonymous spread, revise or hold; consensus map with dissent kept.

InjayTseng/graph-engineering-on-research · 49 tokens

issue-tree

Use on a big fuzzy problem before any research is dispatched — MECE decomposition into a tree whose leaves route onward to research or roundtable rounds by themselves.

InjayTseng/graph-engineering-on-research · 34 tokens

route

Use when you have a question and are not sure which of the six templates applies — classifies it, quotes the cost from measured runs before spending anything, and says so when none of them fits.

InjayTseng/graph-engineering-on-research · 42 tokens

diamond-research

Use when researching unknown territory a question public data can answer — a market, competitors, regulation. Splits into parallel search angles, adversarially verifies every claim, reports with confidence labels.

InjayTseng/graph-engineering-on-research · 42 tokens